- Developer
- Alibaba Tongyi Lab (Alibaba Cloud)
- First released
- February 2025 (Wan2.1)
- Latest major version
- Wan2.2 (July 2025)
- License
- Apache 2.0 (open source)
- Capabilities
- Text-to-video, image-to-video, speech-to-video
- Related
- Seedance, Kling AI, Veo, Sora
- Developer
- Alibaba Tongyi Lab (Alibaba Cloud)
- First released
- February 2025 (Wan2.1)
- Latest major version
- Wan2.2 (July 2025)
- License
- Apache 2.0 (open source)
- Capabilities
- Text-to-video, image-to-video, speech-to-video
- Related
- Seedance, Kling AI, Veo, Sora
Wan (also known as Tongyi Wanxiang) is a family of open-source AI video generation models developed by the Tongyi Lab of Alibaba Cloud, capable of producing short videos from text prompts, still images, and audio. The series gained prominence as the first open-source video generation models to use a Mixture-of-Experts (MoE) architecture, delivering higher quality and efficiency without proportionally higher compute costs.[1][2]
History
Alibaba released Wan2.1 in February 2025, which established the series as a leading open-source option for image-to-video generation.[3] Wan2.2 followed on 28 July 2025, introducing the MoE architecture, expanded training data, and improved aesthetic control.[4] In August 2025, Alibaba released Wan2.2-S2V, a speech-to-video variant that generates digital-human videos from a single portrait photo and an audio clip.[3] By late 2025, the Wan model series had accumulated more than 6.9 million downloads on Hugging Face and ModelScope.[3]
Technology
Wan models are diffusion-based video generators available in multiple variants, including text-to-video (T2V) and image-to-video (I2V) configurations.[2] Wan2.2's flagship 14-billion-parameter MoE models, such as I2V-A14B, generate stable video with reduced unrealistic camera movement at 480p and 720p resolutions, while a lightweight TI2V-5B variant combines text- and image-to-video in a single high-compression model, using a custom Wan2.2 VAE with 64× compression to produce 720p at 24 fps in under nine minutes on a consumer RTX 4090 GPU.[4][5] Compared with Wan2.1, Wan2.2 was trained on 65.6 percent more images and 83.2 percent more videos, and leads both open-source and closed-source benchmarks on Wan-Bench 2.0.[4] The models are released under the Apache 2.0 license and distributed through Hugging Face, GitHub, and Alibaba Cloud's ModelScope platform, with commercial licensing available for enterprise use.[3][4][5]
Applications
Wan is used for cinematic short-form video production, advertising, product visualisation, and digital-human content, including speech-driven avatars that can speak, sing, and perform.[3][5] Because the models are open-weight and run on consumer hardware, they are widely deployed through community tools such as ComfyUI workflows and on rented GPU infrastructure, positioning Wan as a low-cost alternative to closed commercial APIs in the AI video space.[4][5]
>See Also
References
Alibaba Cloud has operated in Malaysia since 2017, opened its third Malaysian data centre in August 2025, and launched a new public cloud region in Johor with two data centres, bringing its total Malaysian footprint to five facilities — its largest in Southeast Asia — with reported plans for a fourth data centre by 2026.[6][7][8] Alibaba Cloud has positioned itself as a strategic partner supporting Malaysia's goal of becoming an AI nation by 2030, with local partnerships including Ryt Bank and PRO-NET.[8] For Malaysian creators and agencies, Wan's open-source availability means video generation can be run on local or regional GPU infrastructure — reducing per-video costs compared with paid APIs such as ByteDance's Seedance — while Alibaba Cloud's Malaysian regions provide lower-latency hosting for Qwen- and Wan-based applications. As with all generative video tools, Malaysian users should observe PDPA requirements when generating content from personal data and follow responsible-AI practices under the National AI Office's governance framework.
References
- ↑[MimicPC: Wan2.2 — Leading Open-Source AI Image to Video Model](https://www.mimicpc.com/learn/wan-22-best-open-source-ai-image-to-video-model)
- ↑[Vast.ai: Wan 2.2 Explained — A New Approach to AI Video Generation](https://vast.ai/article/wan-2-2-explained-new-approach-ai-video-generation)
- ↑[IT Brief Asia: Alibaba unveils Wan2.2-S2V, open-source speech-to-video tool](https://itbrief.asia/story/alibaba-unveils-wan2-2-s2v-open-source-speech-to-video-tool)
- ↑[NYU Shanghai RITS: Wan2.2 — Alibaba's Open-Source Breakthrough in AI Video Generation](https://rits.shanghai.nyu.edu/ai/wan2-2-alibabas-open%E2%80%91source-breakthrough-in-ai-video-generation)
- ↑[Wan2.2 — Official Project Site](https://wan22.io)
- ↑[Alibaba Cloud Community: Alibaba Cloud Expands Global AI Infrastructure with New Data Centers in France, Japan, Malaysia, and Mexico](https://www.alibabacloud.com/blog/alibaba-cloud-expands-global-ai-infrastructure-with-new-data-centers-in-france-japan-malaysia-and-mexico_603278)
- ↑[Alibaba Cloud Global Locations](https://www.alibabacloud.com/en/global-locations?_p_lc=1)
- ↑[W.Media: Alibaba set to launch 4th data center in Malaysia by 2026](https://w.media/alibaba-set-to-launch-4th-data-center-in-malaysia-by-2026)